American Journal of Medical Genetics (Neuropsychiatric Genetics) 67:369-377 (1996)

New Phenotype Definition of Attention Deficit

Hyperactivity Disorder in Relatives for
Genetic Analyses
Sharon Milberger, Stephen V.Faraone, Joseph Biederman, Marcia Testa, and Ming T. Tsuang
Pediatric Psychopharmacology Unit, Massachusetts General Hospital (S.M., S.V.F., J.B.), Harvard Institute of
Psychiatric Epidemiology and Genetics (S.V.F., M.T.T.) and Harvard Schools of Medicine and Public Health
(S.M., S.V.F., J.B., M.T., M.T.T.), Harvard University, Boston, Massachusetts

The goal of the present investigation was to 1990; Stewart et al., 1980; Welner et al., 19771. Al-
create a phenotype definition in relatives of though familial transmission may be of genetic or envi-
probands that reflects a more genetic form ronmental origin, it is believed that genes play an im-
of attention deficit hyperactivity disorder portant role in the familial transmission of ADHD
(ADHD). Logistic regression was applied to based on data from half-sibling, twin, adoption, and
the first-degree relatives of ADHD and nor- segregation analysis studies [Cantwell, 1975; Deutsch
mal control probands to create a quantitative et al., 1990; Faraone and Santangelo, 1992; Goodman
phenotype that combined information across and Stevenson, 1989a,b; Lopez, 1965; Morrison and
psychiatric, cognitive, and demographic do- Stewart, 1973; Rutter et al., 1963; Safer, 1973; Torger-
mains. Models were run separately in moth- sen and Kringlen, 1978; Willerman, 19733. If a single
ers, fathers, sisters, and brothers. Although major gene does exist, misclassification could make it
there was some overlap between the vari- difficult, if not impossible, t o detect with linkage or as-
ables retained in each model, no two models sociation studies.
had exactly the same variables. Our results In this investigation we were interested in classify-
suggest that the use of fitted logits may be ing ADHD into categories that discriminated between
valuable as a potential index of caseness. genetic and nongenetic subtypes of the disorder. This is
Since different characteristics were in- what Tsuang et al. [1993]referred to as “psychiatric ge-
cluded in different groups of relatives, our netic nosology.” This approach does not suggest that
results suggest that gender and generation such a nosology will be useful for clinicians or that it
may moderate the expression of ADHD. should replace DSM-111-R. What it does imply is that
0 1996 Wiley-Liss, Inc. psychiatric genetics need not rely on diagnostic con-
structs for other purposes [Tsuang et al., 19931. A psy-
KEY WORDS: psychiatric morbidity, famil- chiatric genetic nosology needs to address the key mea-
ial transmission, logistic re- surement issues in psychiatric genetics. Of primary
gression concern is the group of false-positives. This refers to a
group of subjects who are classified as having ADHD
but are incorrectly classified as having a genetic sub-
INTRODUCTION form of the illness (i.e., subjects who have the pheno-
type but not the genotype). These false-positives are of
Family studies of attention deficit hyperactivity dis- critical concern in genetic studies since they may lead
order (ADHD) have found that the relatives of ADHD to reduced statistical power and attenuated evidence
children are at high risk for ADHD [Biederman et al., for genetic linkage [Chen et al., 1992; Tsuang et al.,
1990, 1991a,b; Cantwell, 1972; Faraone et al., 1992; 19931.False-negatives (i.e., subjects who do not express
Faraone and Santangelo, 1992; Lahey et al., 1989; the phenotype but have the genotype) are also of con-
Mannuzza and Gittelman, 1984; Morrison, 1980; Mor- cern in genetic analyses, but to a lesser extent since
rison and Stewart, 1971; Schachar and Wachsmuth, they do not have the same impact on statistical power.
Typically, the phenotype definition for ADHD has
been based on the Diagnostic and Statistical Manual of
Mental Disorders (DSM) classification system, which
Received for publication August 2, 1995; revision received
December 2, 1995.
treats diagnoses as dichotomous traits [American Psy-
chiatric Association, 1987; Tsuang et al., 19931. In
Address reprint requests to Dr. Joseph Biederman, Pediatric
Psychopharmacology Unit, ACC-725, Massachusetts General many cases, however, such a dichotomy comes about af-
Hospital, 15 Parkman Street, Boston, MA 02114. ter examining several variables on an individual, and
0 1996 Wiley-Liss, Inc.
370 Milberger et al.
then applying a rule as t o who is affected and who is un-
affected. This dichotomization process can yield consid-
erable loss of information and misclassification. Ott
[1991] developed an approach that addressed these is-
sues. His scheme is based on a measure of certainty
that an individual is truly affected and assigns weights
according to the probability that the subject is affected.
These weights can be the predictions made by mathe-
matical models of disease expression [Tsuang et al.,
19931. In a linkage analysis, these weights can be used
to quantify the penetrance of each genotype for diag-
noses made with varying degrees of certainty [Ott,
19911.
Therefore, differences in demographic and clinical
features in relatives of ADHD probands and relatives of
normal controls may be combined to create Ott’s mea-
sure of certainty that would weight the contribution of
individual relatives in a linkage analysis by the chance
that they have the genetic form of ADHD. If any one of
these characteristics were examined individually it
might not be highly predictive, whereas the combina-
tion of a number of characteristics into a composite pro-
file might yield considerable predictive ability [Blacker
et al., 1993; Blacker and Tsuang, 19931.
Specific characteristics that may be of particular use
in the new phenotype definition in relatives are gender
and comorbidity status. Gender is relevant in that it
may moderate the phenotypic expression of ADHD. For
example, males may be more hyperactive and impul-
sive while females may be more inattentive and have
more internalizing problems. This gender difference is
evidenced by ADHD being 6-9 times more prevalent in
males than females [Anderson et al., 1987; Bird et al.,
1988; Safer and Krager, 19881.
Comorbidity status may be another important char-
acteristic to examine in the new phenotype definition,
since it is believed that ADHD with various comorbid
psychiatric disorders represents a group of conditions
with different etiological risk factors rather than a sin-
gle homogeneous clinical entity. More specifically, it is
believed that ADHD with conduct disorder may repre-
sent a distinct genetic subtype ofADHD [Faraone et al.,
19911,that ADHD and major depression share common
familial etiologic factors [Biederman et al., 1991b1, and
that ADHD and anxiety disorders are etiologically in-
dependent [Biederman et al., 1991al.
Modeled after recent work by Blacker and Tsuang
[19931 and by Faraone et al. [19951, the present work
uses logistic regression to discriminate relatives of
ADHD probands from relatives of normal controls. As
discussed in Subjects and Methods, this allowed us t o
~ ~
TABLE I. Number of First-Dem-ee Biological Relatives of Probands
~ ~
Relatives of ADHD probands
Type of Total number Number included
relative (N = 454) (% of total)
Mothers 140 136 (97%)
Fathers 140 111(79%)
Sisters 81 68 (84%)
Brothers 93 78 (84%)
create a quantitative phenotype definition in relatives
that combined information across psychiatric, cogni-
tive, and demographic domains.
SUBJECTS AND METHODS
Subjects
We analyzed data from a family-genetic study of
ADHD that has been described previously [Biederman
et al., 19921. Briefly, this study selected two groups of
index children: 140 ADHD probands and 120 normal
controls. These groups had 454 and 368 first-degree
biological relatives, respectively. The number of moth-
ers, fathers, sisters, and brothers in each of these
groups is delineated in Table I. We sampled families
through Caucasian, non-Hispanic male probands between
ages 6-17 years. Potential probands were excluded if
they had been adopted, or if their nuclear family was
not available for study. We also excluded probands if
they had major sensorimotor handicaps (e.g., paralysis,
deafness, or blindness), psychosis, autism, or a full-scale
I& 4 0 . Subjects from the lowest socioeconomic class
(SES VI) [Hollingshead, 19751 were excluded t o mini-
mize the potential confounding of social adversity. All
ADHD probands met DSM-111-R diagnostic criteria for
current ADHD at time of clinical referral. Two inde-
pendent sources provided the index children, as de-
scribed elsewhere [Biederman et al., 19921. A three-
stage ascertainment procedure was used to select all
probands regardless of their source of referral [i.e.,
Massachusetts General Hospital (MGH) or Harvard
Community Health Plan (HCHP)]. In each case, sam-
pling was based on consecutive referrals.
For purposes of this investigation, only those parents
who were interviewed directly about themselves were
included in the analysis. This led to the exclusion of 42
(16%)fathers and 2 (1%)mothers. Moreover, relatives
with incomplete data were also excluded. Six (2%)
mothers, 10 (4%)fathers, 17 (11%) sisters, and 22 (14%)
brothers were excluded for this reason. Thus, data on
252 (97%) mothers, 207 (80%) fathers, 131 (89%) sis-
ters, and 133 (86%)brothers were used in the analysis.
Table I provides a breakdown of the number of avail-
able subjects by whether they were related to an ADHD
proband or to a normal control proband.
Procedures
All diagnostic assessments were made using DSM-
111-R-based structured interviews. Psychiatric assess-
ments of probands and siblings were made with the
Kiddie SADS-E (EpidemiologicVersion) [Orvaschel and
Puig-Antich, 19871. Diagnoses were based on indepen-
Relatives of normal control probands
Total number Number included
(N = 368) (% of total)
120 116 (97%)
119 96 (81%)
67 63 (94%)
62 55 (89%)

dent interviews with mothers and direct interviews of
probands and siblings, except for children below age 12
years, who were not directly interviewed. Diagnostic
assessments of parents were based on direct interviews
with each parent, using the Structured Clinical Inter-
view for DSM-111-R (SCID) [Spitzer and Williams, 19861.
All assessments were made by raters who were blind
to proband diagnosis (ADHD or control) and ascertain-
ment site [MGH or Health Maintenance Organization
(HMO)]. Mothers’ interviews about their children were
sequenced after the direct interview with the mother
about herself had been completed. Different raters con-
ducted the direct interviews of siblings and the indirect
interviews with mothers about their children. Inter-
view data were collected on all siblings in both ADHD
and control families. All parents signed a written con-
sent form for participation in the study.
Interviews were conducted by five raters with under-
graduate degrees in psychology who had been trained
to high levels of interrater reliability. Kappa coeffi-
cients of agreement were computed between raters and
three experienced, board-certified child and adult psy-
chiatrists who listened to audiotaped interviews made
by the raters. A kappa of 1.0 was obtained for ADHD
(95% confidence interval, 0.8-1.0).
Diagnoses were considered positive if, based on in-
terview results, DSM-111-R criteria were unequivocally
met. All diagnostic uncertainties were resolved by a
committee of four board-certified child and adult psy-
chiatrists who were blind t o the subject’s ascertain-
ment group, ascertainment site, all data collected from
other family members, and all nondiagnostic data (e.g.,
cognitive functioning). Diagnoses presented for review
were considered positive only if a consensus was
achieved that criteria were met to a degree that would
be considered clinically meaningful. For children older
than 12 years, symptom data from direct and indirect
interviews were combined by considering a symptom
positive if it were endorsed in either interview. Chil-
dren below age 12 years were not directly interviewed
for a number of reasons: they have limited language ca-
pabilities, are often unable to map events over time,
and have limited abilities with abstract concepts. Given
these shortcomings, there is a real issue as to whether
the child’s self-perceptions, memories, emotions, and
reported behavior can be reliably assessed through self-
report. Studies on the use of interview techniques
among children under age 12 years show that they only
understand between 28-38% of the questions [Breton
et al., 19953, that their replies are not reliable [Achen-
bach et al., 19871, and that their parents tend to
be more reliable informants [Edelbrock et al., 1986;
Gutterman et al., 19871.
In parents and siblings, symptom scores were cre-
ated corresponding to ADHD and the various psy-
chiatric disorders that are frequently comorbid with
ADHD (e.g., conduct disorder, major depression, and
anxiety disorders). That is, for each of these diagnostic
categories, the number of symptoms endorsed was
summed to give an index of severity, instead of using
the dichotomous DSM-111-R classification system.
Academic achievement was assessed with the Arith-
metic Subtest of the Wide Range Achievement Test
New Phenotype Definition of ADHD 371
(WRAT-R) [Jastak and Jastak, 19851 and the Gilmore
Oral Reading Test [Gilmore and Gilmore, 19681. Intel-
lectual functioning was assessed with the vocabulary,
block design, digit span, and digit symbol subtests of
the Wechsler Intelligence Scale for Children-Revised
(WISC-R) [Wechsler, 19741. A total cognitive score was
created by standardizing each of the above-mentioned
cognitive variables and summing them. This cognitive
score was then itself standardized to increase the in-
terpretability of this measure.
Data Analysis
Univariate comparisons were conducted to assess the
ability of individual variables to discriminate relatives
of ADHD probands from relatives of normal controls.
Chi-square tests were used for categorical variables,
two sample t-tests were used for approximately normal
continuous or ordinal variables, and Wilcoxon tests
were used for continuous or ordinal variables with
skewed distributions.
In an attempt to define a phenotype of ADHD that is
highly prevalent among relatives of ADHD probands
but rare in the relatives of normal controls, logistic re-
gression was applied to the first-degree relatives of
ADHD and normal control probands. Let P denote the
probability of being a first-degree relative of an ADHD
proband. The logarithm of the odds, log[P/(l - PI], is
called the logit of P. Logistic regression models the logit
as a linear function of the explanatory variables, which
included the symptom scores of various psychiatric dis-
orders, the cognitive summary functioning score, so-
cioeconomic status, and past and current global assess-
ment of functioning scores. All explanatory variables
were submitted to a forward stepwise logistic regres-
sion using a significance level t o enter or stay in the
model of 0.05. At each step, the improvement in X2-test
was used to check whether the variable entered at that
step significantly improved the log-likelihood.Parame-
ter estimates were obtained using maximum likeli-
hood. After the last step, the hypothesis that the logis-
tic model fit the data adequately was tested with the
Hosmer-Lemeshowtest [Hosmer and Lemeshow, 19891.
Once the final logistic regression models were deter-
mined in each group of relatives, the equivalence of
these models was tested. This was done by using iden-
tical variables in each of the four models (a variable
was included if it was retained in any of the stepwise
logistic regression models). The log-likelihoods from
these four individual models were then summed and
compared with the log-likelihood from the model
looking a t all relatives a t once. The difference in log-
likelihoods was multiplied by two and evaluated using
a X2-test.
After estimates of the coefficients were obtained, an
estimate of the probability of being a relative of an
ADHD proband was calculated for each relative in the
study as follows:
eOoat(i))l
P = [1 + e(logit(P))
1
These predicted probabilities provided a quantitative
index of how likely a subject was t o be a relative of an
372 Milberger et al.
ADHD proband based on the relative’s profile of char- two models looking a t siblings (i.e., a proband could
acteristics as retained in the logistic regression model. have had two or more same-sex siblings). Since our pri-
An important issue to consider when working with mary aim was the separation of relatives of ADHD and
family data is that family members are not statistically normal control probands rather than hypothesis-
independent of one another due to shared cultural testing, we ignored this lack of independence in the
andor genetic factors. Although this dependency is not models using sisters and brothers. However, the mag-
likely t o have a large effect on parameter estimates it nitude of this problem should not be great, since the
can influence statistical testing. Dependency between number of brothers and sisters in these families was
subjects results in decreased variability which, in turn, not large. More specifically, in those families with
can yield underestimates of significance levels (i.e., brothers, 84% had only 1brother, 13%had 2 brothers,
smaller P values). To control for this, four different lo- and 3%had 3 brothers. Similarly, in those families that
gistic models were run for each type of relative of had sisters, 80% had only 1 sister, 17% had 2 sisters,
probands: one for mothers, one for fathers, one for sis- and 3% had 3 sisters.
ters, and one for brothers. This attempt to eliminate de-
pendency also helped address the issues of gender and RESULTS
generational effects, which may have considerable im- Table I1 presents the associations between group sta-
pact on ADHD family data. However, it was still possi- tus (i.e., being a relative of an ADHD proband vs. a rel-
ble that there were some related family members in the ative of a normal control proband) and each of the

TABLE IIa. Univariate Analyses in Mothers (N = 252)

Mothers of ADHD Mothers of normal
probands (N = 136) controls (N = 116) P value
Extra help in school 25% (34) 9% (11) 0.002
Repeated grade in school 20% (27) 5%(6) 0.0006
Special class in school 1%(1) 0% ( 0 ) 1.0
Mean t SD Mean t SD
ADHD symptom score 2.4 -+ 3.5 0.72 t 1.5 0.0001
Conduct disorder symptom score 0.45 f 0.86 0.20 2 0.51 0.005
Major depression symptom score 3.6 f 3.3 2.0 t 2.9 0.0001
Separation anxiety symptom score 0.50 f 1.2 0.36 t 0.77 0.10
Panic disorder symptom score 1.5 ? 1.5 0.43 t 1.4 0.0002
Antisocial personality disorder 1.6 f 1.1 1.3 t 0.63 0.006
symptom score
Generalized anxiety disorder 2.3 -+ 3.8 0.5 t 1.9 0.0001
symptom score
Cognitive summary score -1.2 t 3.7 0.65 t 2.9 0.0001
Current global assessment of 69.3 t 10.1 75.9 t 7.0 0.0001
functioning score
Past global assessment of 57.9 t 13.2 66.9 t 11.1 0.0001
functioning score
Socioeconomic status 1.8 f 0.83 1.5 2 0.67 0.0007
TABLE IIb. Univariate Analyses in Fathers (N = 207)
Fathers of ADHD Fathers of normal
Drobands (N = 111) controls (N = 96) P value
Extra help in school 26% (29) 15% (14) 0.06
Repeated grade in school 23% (25) 21% (22) 0.87
Special class in school 3%(3) 2% (2) 1.0
Mean t SD Mean t SD
ADHD symptom score 3.8 t 3.8 1.7 f 2.4 0.0001
Conduct disorder symptom score 1.2 t 1.5 0.76 t 1.5 0.006
Major depression symptom score 3.0 t 2.9 1.8 t 2.6 0.003
Separation anxiety symptom score 0.39 t 0.75 0.19 ? 0.53 0.005
Panic disorder symptom score 0.52 2 1.7 0.44 t 1.5 0.60
Antisocial personality disorder 2.1 t 1.3 1.6 t 0.93 0.005
symptom score
Generalized anxiety disorder 1.4 t 3.2 0.98 t 2.2 0.71
symptom score
Cognitive summary score -0.29 f 4.7 1.3 f 3.7 0.01
Current global assessment of 68.2 t 10.2 72.6 t 8.0 0.001
functioning score
Past global assessment of 56.5 ? 12.5 62.9 t 11.9 0.0001
functioning score
Socioeconomic status 1.6 t 0.79 1.4 t 0.59 0.06
 New Phenotype Definition of ADHD 373
TABLE IIc. Univariate Analyses in Sisters (N = 131)
Sisters of ADHD Sisters of normal
probands (N = 68) controls (N = 63) P value
Extra help in school 35% (24) 22% (14) 0.12
Repeated grade in school 0% (0) 0% (0) NIA
Special class in school 3%(2) 10%(8) 0.17
Mean 2 SD Mean ? SD
ADHD symptom score 2.9 t 3.8 1.5 2 2.6 0.04
Agoraphobia symptom score 0.21 ? 0.59 0.13 ? 0.42 0.66
Conduct disorder symptom score 0.59 ? 1.2 0.22 ? 0.63 0.03
Major depression symptom score 2.1 t 3.1 2.2 t 3.0 0.87
Separation anxiety symptom score 1.1t 1.5 0.78 t 1.0 0.23
Simple phobia symptom score 1.8 ? 2.0 1.2 ? 1.8 0.06
Social phobia symptom score 1.3 t 1.9 0.71 ? 1.4 0.08
Overanxious symptom score 1.6 t 1.9 1.4 ? 1.8 0.32
Panic disorder symptom score 0.53 ? 1.5 0.55 ? 1.6 0.71
Cognitive summary score -0.29 t 3.3 0.44 2 3.0 0.19
Current global assessment of 70.3 2 11.0 72.3 ? 8.7 0.22
functioning score
Past global assessment of 63.8 2 12.5 66.3 2 11.4 0.19
functioning score
Socioeconomic status 1.7 t 0.80 1.5 2 0.70 0.23
TABLE IId. Univariate Analyses in Brothers (N = 133)
Brothers of ADHD Brothers of normal
probands (N = 78) controls (N = 55) P value
Extra help in school 41% (32) 29% (16) 0.20
Repeated grade in school 1%(1) 0% (0) 0.59
Special class in school 10% (8) 11% (6) 0.96
Mean ? SD Mean 2 SD
ADHD symptom score 3.9 ? 4.2 2.3 t 2.9 0.05
Agoraphobia symptom score 0.16 2 0.51 0.22 ? 0.41 0.11
Conduct disorder symptom score 1.0 ? 1.5 0.55 ? 1.2 0.03
Major depression symptom score 2.1 ? 3.0 1.4 ? 2.4 0.33
Separation anxiety symptom score 1.2 2 1.5 0.69 ? 0.79 0.08
Simple phobia symptom score 1.4 t 1.8 1.12 1.7 0.40
Social phobia symptom score 0.78 ? 1.6 0.58 ? 1.2 0.57
Overanxious symptom score 1.6 ? 1.6 0.95 ? 1.3 0.01
Panic disorder symptom score 0.51 t 1.8 0.27 2 1.2 0.31
Cognitive summary score -0.96 ? 3.7 1.2 2 3.8 0.003
Current global assessment of 67.4 ? 10.6 70.1 ? 8.4 0.07
functioning score
Past global assessment of 60.6 t 11.8 66.0 2 10.6 0.006
functioning score
Socioeconomic status 1.8 2 0.84 1.4 ? 0.63 0.002

explanatory variables t h a t was entered into the step-
wise logistic regression models. Mothers, fathers, sis-
ters, and brothers were examined separately (Table
IIa-d, respectively). A number of significant differences
were seen between mothers of ADHD probands and
mothers of normal controls (Table IIa). Compared with
mothers of normal controls, mothers of ADHD
probands had higher rates of extra help and repeated
grades in school; they also had higher ADHD, conduct
disorder, major depression, panic disorder, antisocial
personality disorder, and generalized anxiety disorder
symptom scores, lower current global assessment of
functioning, past global assessment of functioning, cog-
nitive, and socioeconomic status scores.
A number of significant differences were also found
between fathers of ADHD probands and fathers of nor-
mal control probands (Table IIb). The fathers of ADHD
probands had higher ADHD, conduct disorder, major
depression, separation anxiety, and antisocial person-
ality disorder symptom scores, lower cognitive, current
and past global assessment of functioning scores, and
lower socioeconomic status compared with fathers of
normal controls.
Only two significant differences were found between
sisters of ADHD probands and sisters of normal con-
trols (Table IIc). Sisters of ADHD probands had higher
ADHD and conduct disorder symptom scores than sis-
ters of normal controls.
A number of significant differences were seen be-
tween brothers of ADHD probands and brothers of nor-
mal controls (Table IId). Brothers of ADHD probands
had higher ADHD, conduct disorder, and overanxious
disorder symptom scores, lower cognitive and past glo-
bal assessment of functioning scores, and lower socio-
economic status than brothers of normal controls.
Table I11 presents the final logistic regression models
in the relatives. Models were run separately in moth-
ers, fathers, sisters, and brothers (Table IIIa-d, respec-
tively). Although there was some overlap between the
variables that were retained in each model, no two
374 Milberger et al.
TABLE 111. Logistic Regression Models in Relatives
a. In mothers*
Variable Odds ratio Standard error Z P value
ADHD symptom score 1.21 0.09 2.5 0.012
Cognitive summary score 0.87 0.04 -2.9 0.003
Current global assessment of functioning score 0.95 0.02 -2.5 0.013
Generalized anxiety disorder symptom score 1.17 0.08 2.3 0.022
* Hosmer-Lemeshow goodness of fit test: x2 = 274.72, df = 244, P = 0.10.
b. In fathers*
Variable Odds ratio Standard error Z P value
ADHD symptom score 1.20 0.06 3.5 0.001
Major depression symptom score 1.13 0.06 2.2 0.030
Socioeconomic status 1.54 0.34 2.0 0.047
* Hosmer-Lemeshow goodness of fit test: x2 = 118.40, df = 111, P = 0.30.
c. In sisters*
Variable Odds ratio Standard error Z P value
ADHD symptom score 1.14 0.07 2.2 0.030
* Hosmer-Lemeshow goodness of fit test: $ = 20.96, df = 13, P = 0.07.

d. In brothers"
Variable Odds ratio Standard error Z P value
Cognitive summary score 0.86 0.04 -2.9 0.004
Past global assessment of functioning score 0.96 0.02 -2.4 0.018
* Hosmer-Lemeshow goodness of fit test: x2 = 132.99, df = 130, P = 0.41.

models had exactly the same variables. For example,
the logistic regression model discriminating mothers of
ADHD probands from mothers of normal controls
showed that the ADHD symptom score (i.e., number of
ADHD symptoms that were endorsed), the cognitive
summary score, the current global assessment of func-
tioning score, and the generalized anxiety disorder
symptom score were the variables in the profile t h a t
yielded the greatest discriminating ability (Table IIIa).
Similarly, the logistic regression model in fathers also
showed that the ADHD symptom score was a n impor-
tant characteristic in the profile that discriminated fa-
thers of ADHD probands from fathers of normal control
probands (Table IIIb). In contrast, the model in fathers
showed that the major depression symptom score and
socioeconomic status were also part of the discriminat-
ing profile, while the cognitive summary score, the cur-
rent global assessment of functioning score, and the
generalized anxiety disorder symptom score were not.
The logistic regression model in sisters was similar to
the models seen in both parent groups in that the
ADHD symptom score was a n important characteristic
in discriminating group status, In contrast to the mod-
els in parents, the profile that yielded the greatest dis-
criminating ability in the sisters consisted of only t h a t
one variable (Table IIIc). The logistic regression model
in the brothers differed from the models in all the other
relative groups in that the ADHD symptom score was
not part of the profile discriminating group status
(Table IIId). Moreover, the profile in brothers included
the past global assessment of functioning score, which
was not seen in any other relative profiles. Similar to
the model in mothers, the cognitive summary score
was also a n important characteristic of the profile in
brothers.
The equivalence of these four logistic regression mod-
els was evaluated by comparing the sum of the likeli-
hoods from the four separate models and the likelihood
from the model examining all relatives a t once. The re-
sults of this analysis yielded a x2 = 25.2 on 21 degrees
of freedom ( d f )( P > 0.05).
DISCUSSION
Using logistic regression models, we identified a
quantitative phenotype in relatives that consists of
measures across psychiatric, cognitive, and psychoso-
cia1 domains. I t is possible that psychiatric genetics
may be better served by diagnostic constructs created
for that particular intent than by existing diagnostic
criteria that currently exist for broader purposes in
clinical settings. The finding t h a t so many of the uni-
variate analyses were significant, especially in parents,
is consistent with what we expected: that these fea-
tures may reflect the phenomenology of a n ADHD gene.
However, there is a lot of redundancy among these fea-
tures and therefore when they were entered simultane-
ously into a logistic regression model only some of them
were retained.
The combination of these variables in a quantitative
logistic regression model yielded a discriminating ge-
netic phenotype for mothers of children with ADHD
that included four variables: the ADHD symptom score,

the cognitive summary score, the current global assess-
ment of functioning score, and the generalized anxiety
symptom score. In contrast, the combination of these
variables yielded a genetic phenotype for the fathers
that included three variables: the ADHD symptom
score, the major depression symptom score, and socio-
economic status. That cognitive scores were a significant
discriminator in mothers but not in fathers is not sur-
prising, since impaired cognitive functioning is more
prevalent in males than females. This may be related to
the tenet that males are more prone t o central nervous
system insults [Hynd and Semrud-Clikeman, 1989a,b;
O'Callaghan et al., 19921.Analogously, it is not surpris-
ing that the total number of major depression symp-
toms (summary score) was a significant discriminator
in fathers but not in mothers, since major depression is
estimated to be twice as prevalent in females than in
males. The finding that socioeconomic status (SES) is
retained in the father model may be indicative that
SES is a correlate of the higher rates of psychopathol-
ogy in fathers of ADHD probands. Alternatively, it
could be indicative that SES is itself a causal factor.
Further work is needed t o determine the influence of
SES. The differences between mothers and fathers
highlight the importance of considering demographic
factors such as gender and SES in the definition of fa-
milial phenotypes.
It is difficult to interpret the findings from the logis-
tic regression model in sisters, since the model did not
fit the data well (Hosmer-Lemeshow Goodness of Fit
Test x2 = 20.96, df = 13, P = 0.07). This poor fit sug-
gests that the DSM-111-R categorical approach is supe-
rior in looking a t the phenotype in sisters of probands
with ADHD.
The combination of variables in brothers of children
with ADHD yielded a genetic phenotype that included
the following two variables: cognitive functioning sum-
mary score, and past global assessment of functioning
score. The finding that the number of ADHD symptoms
in boys was not part of the profile discriminating among
brothers is a t first glance puzzling. Although a signifi-
cant difference in ADHD symptom score was observed
between brothers of ADHD probands and brothers of
normal control probands on a univariate level (odds ra-
tio = 1.13, P = 0.021, the ADHD symptom score did not
add any discriminating ability to the profile. This is
probably due, in part, to the fact that ADHD was so
prevalent in brothers of normal controls (9%). More-
over, the ADHD symptom score was significantly corre-
lated with the past global assessment of functioning
score (Spearman correlation coefficient = -0.52, P =
0.0001) and slightly correlated with cognitive function-
ing summary score (Spearman correlation coefficient =
-0.16, P = 0.07).
The finding that the discriminating variables in the
brothers were reflective of socioeconomic status is par-
ticularly interesting. Since these variables could be as-
sociated with all types of psychopathology they may not
necessarily be useful in developing a discriminating ge-
netic taxonomy for ADHD. The absence of SES differ-
ences in the sisters is consistent with existing litera-
ture suggesting that males (i.e.)brothers) may be more
New Phenotype Definition of ADHD 375
affected by environmental factors such as perinatal
complications while females (i.e., sisters) may repre-
sent a more genetic group. Additional study is needed
in order to fully evaluate this hypothesis.
While our findings in the four groups of relatives sug-
gest that we were successful in creating a quantitative
phenotype in adults (i.e., mothers and fathers), our ap-
proach was not as useful in children (i.e., sisters and
brothers). The fact that our approach did not improve
the ability to discriminate among children is not sur-
prising, since the diagnostic criteria have been de-
signed for this age group and, in general, have been
shown to be useful. While we believe our new quantita-
tive phenotype in adults may be useful in the context of
genetic analyses, we do not imply that such a nosology
will be useful for clinicians or that it should replace
DSM-111-R.
Our work must be interpreted in the context of its
methodological limitations. An issue in this study is re-
call bias, which may be especially problematic in the as-
sessment of childhood diagnoses in adult relatives,
since it requires them to recall experiences from child-
hood. For example, the diagnosis of ADHD requires an
onset of symptoms by age 7 years. Moreover, another
limitation of our work is that the use of maternal re-
ports to make diagnoses of children may have led to
those diagnoses being influenced by maternal psy-
chopathology. However, incorporation of maternal re-
ports is standard clinical practice, since children are
considered to be poor reporters of their symptoms
[Achenbach et al., 19871.
Another potential limitation in clinical studies is
Berkson's bias, i.e., the sample may include a mislead-
ingly high proportion of subjects with multiple diag-
noses just because referral will have been influenced by
the occurrence of each separate condition [Berkson,
19461. However, high levels of comorbidity have been
observed in both clinical [Biederman et al., 1987,
1991a,b, 19921and epidemiological samples [Anderson
et al., 1987; Angold and Costello, 1993; Bird et al.,
19881, suggesting that the high levels of comorbidity
seen in clinical samples are not solely an artifact of
Berkson's bias. Similarly, since 50% of the referrals to
the Pediatric Psychopharmacology Unit a t Massachu-
setts General Hospital have never been evaluated or
treated, this sample does not have the severity bias
that might be associated with a tertiary care facility.
Since we did not control for age, and since the original
sample of probands in this investigation came from
clinical referrals, we do not know to what degree these
findings will generalize to nonreferred populations in
the community.
An additional potential shortcoming is that not all
relatives were included in the analysis due to indirect
interviewing (mostly of fathers) or inadequate data.
The most extreme situation was the case of fathers of
ADHD probands, where only 79% were available for in-
clusion in the analysis. However, as shown in Table I,
no differences were noted in availability of subjects,
whether they were a relative of an ADHD proband or a
relative of a normal control.
Another pitfall of this investigation is that we were
not able to determine whether the different predictive
376 Milberger et al.
models in the four groups of relatives were significantly
different from one another. Although our statistical test
comparing the four models did not reveal statistically
significant differences, this could be due to insufficient
power. Therefore, our conclusions can only be con-
sidered tentative. Further studies of larger size are
needed before more definitive conclusions can be made.
A further limitation of this investigation is that our
data can only demonstrate a familial association be-
tween the profile of characteristics in the relatives and
ADHD in the proband. We infer that it is reasonable to
use the familial form of ADHD as a proxy for the ge-
netic form, since it is believed that genetic factors me-
diate the familial transmission of ADHD. However,
studies of twin or adoption samples are needed to ver-
ify that genetic factors account for the familial trans-
mission of the profile of characteristics seen in the rel-
atives. Since we do not yet know the nature of true
genetic cases, familial aggregation may be a reasonable
standard for the development of phenotype definitions
for use in genetic analyses [Blacker et al., 19931.
Another limitation is that by extending our pheno-
type definition t o include anxiety in mothers and major
depression in fathers, it is possible that the false-
positive rate could have been increased. Moreover, the
models used may be incorrect or based on erroneous as-
sumptions. The model assumes that an ADHD geno-
type exists. Even if differences between relatives of
ADHD probands and relatives of controls are found,
other more complex genetic and environmental factors
may be responsible for them [Blacker and Tsuang,
19931. However, a phenotype definition based on these
differences should still help eludicate features that sug-
gest familial transmission, and might be of benefit in
weighting the contribution of individual relatives in a
linkage analysis.
Despite these limitations, our results suggest that
the use of fitted logits (or predicted probabilities) may
be valuable as a potential index of caseness for linkage
studies. Since different characteristics were included in
the different groups of relatives, our results are consis-
tent with an existing body of literature [Anderson et al.,
1987; Bird et al., 1988; Safer and Krager, 19881,and in-
dicate that gender and generation (parent vs. sibling)
may moderate expression of ADHD. Therefore, different
phenotype definitions for each group of relatives may
be necessary to delineate this variable expressivity.
ACKNOWLEDGMENTS
This work was supported, in part, by United States
Public Health Service Grant (National Institute of
Mental Health) grant R01 MH-41314-01A2 (to J.B.).
We thank Cheryl Ouellette, Janice Jones, and Mary
Hatch for their contribution to this work.
REFERENCES
Achenbach TM, McConaughy SH, Howell CT (1987): Childladolescent
behavioral and emotional problems: Implications of cross-informant
correlations for situational specificity. Psychol Bull 101:213-232.
Anderson JC, Williams S, McGee R, Silva PA (1987): DSM-I11 disor-
ders in preadolescent children: Prevalence in a large sample from
the general population. Arch Gen Psychiatry 4459-76.
Angold A, Costello E J (1993): Depressive comorbidity in children and
adolescents: Empirical, theoretical and methodological issues. Am
J Psychiatry 150:1779-1791.
American Psychiatric Association (1987): “DSM-111-R.” Washington,
DC: American Psychiatric Association.
Berkson J (1946): Limitations of the application of fourfold table
analysis to hospital data. Biometric Bulletin 2:47-52.
Biederman J, Munir K, Knee D (1987): Conduct and oppositional dis-
order in clinically referred children with attention deficit disorder:
A controlled family study. J Am Acad Child Adolesc Psychiatry 26:
724-727.
Biederman J, Faraone SV, Keenan K, Knee D, Tsuang MT (1990):
Family-genetic and psychosocial risk factors in DSM-I11 attention
deficit disorder. J Am Acad Child Adolesc Psychiatry 29:52&533.
Biederman J , Faraone SV, Keenan K, Steingard R, Tsuang MT
(1991a): Familial association between attention deficit disorder
and anxiety disorders. Am J Psychiatry 148:251-256.
Biederman J , Faraone SV, Keenan K, Tsuang MT (1991b): Evidence
of familial association between attention deficit disorder and ma-
jor affective disorders. Arch Gen Psychiatry 48:633-642.
Biederman J , Faraone SV, Keenan K, Benjamin J, Krifcher B, Moore
C, Sprich S, Ugaglia K, Jellinek MS, Steingard R, Spencer T, Nor-
man D, Kolodny R, Kraus I, Perrin J, Keller MB, Tsuang MT
(1992): Further evidence for family-genetic risk factors in atten-
tion deficit hyperactivity disorder (ADHD): Patterns of comorbid-
ity in probands and relatives in psychiatrically and pediatrically
referred samples. Arch Gen Psychiatry 49:728-738.
Bird HR, Canino G , Rubio-Stipec M, Gould MS, Ribera J, Sesman M,
Woodbury M, Huertas-Goldman S, Pagan A, Sanchez-Lacay A,
Moscoso M (1988): Estimates of the prevalence of childhood
maladjustment in a community survey in Puerto Rico. Arch Gen
Psychiatry 45:1120-1126.
Blacker D, Tsuang MT (1993): Unipolar relatives in bipolar pedigrees:
Are they bipolar? Psychiatr Genet 35-16,
Blacker D, Lavori PW, Faraone SV, Tsuang MT (1993): Unipolar rel-
atives in bipolar pedigrees: A search for indicators of underlying
bipolarity. Am J Med Genet (Neuropsychiatr Genet) 48:192-199.
Breton J, Bergeron L, Valla J, Lepine S, Houde L, Gaudet N (1995):
Do children aged 9 through 11years understand the DISC Version
2.25 questions? J Am Acad Child Adolesc Psychiatry 34:946-956.
Cantwell DP (1972): Psychiatric illness in the families of hyperactive
children. Arch Gen Psychiatry 27:414-417.
Cantwell DP (1975): Genetics of hyperactivity. J Child Psychol Psy-
chiatry 16:261-264.
Chen WJ, Faraone SV, Tsuang MT (1992): Linkage studies of schizo-
phrenia: A simulation study of statistical power. Genet Epidemiol
9:123-139.
Deutsch CK, Matthysse S, Swanson JM, Farkas LG (1990): Genetic la-
tent structure analysis of dysmorphology in attention deficit disor-
der. J Am Acad Child Adolesc Psychiatry 29:189-194.
Edelbrock C, Costello M, Dulcan M, Conover NC, Kala R (1986): Parent-
child agreement on child psychiatric symptoms assessed via struc-
tured interview. J Child Psychol Psychiat 27:181-199.
Faraone SV, Biederman J, Keenan K, Tsuang MT (1991): Separation
of DSM-I11 attention deficit disorder and conduct disorder: Evidence
from a family-genetic study of American child psychiatric patients.
Psychol Med 21:109-121.
Faraone SV, Biederman J , Chen WJ, Krifcher B, Keenan K, Moore C,
Sprich S, Tsuang M (1992): Segregation analysis of attention
deficit hyperactivity disorder: Evidence for single gene transmis-
sion. Psychiatr Genet 2:257-275.
Faraone SV, Santangelo S (1992):Methods in genetic epidemiology. In
Fava M, Rosenbaum JF (eds): “Research Designs and Methods in
Psychiatry.” Amsterdam: Elsevier, pp 93-118.
Faraone SV, Seidman U,Kremen WS, Pepple JR, Lyons MJ, Tsuang
MT (1995): Neuropsychological functioning among the non-
psychotic relatives of schizophrenic patients: A diagnostic efficiency
analysis. J Abnormal Psychology 104:28&304.
Gilmore JV,Gilmore EC (1968): “Gilmore Oral Reading Test.” New
York Harcourt, Brace &World, Inc., pp 1-29.
Goodman R, Stevenson J (1989a): A twin study of hyperactivity I:
An examination of hyperactivity scores and categories derived

from Rutter teacher and parent questionnaires. J Child Psychol
Psychiatry 30:671-689.
Goodman R, Stevenson J (1989b):A twin study of hyperactivity 11:The
aetiological role of genes, family relationships and perinatal adver-
sity. J Child Psychol Psychiatry 30591-709.
Gutterman EM, O’Brien JD, Young G (1987): Structured diagnostic
interviews for children and adolescents: Current status and future
directions. J Am Acad Child Adolesc Psychiatry 23:621-630.
Hollingshead AB (1975): “Four Factor Index of Social Status.” New
Haven: Yale University, Department of Sociology.
Hosmer DW, Lemeshow S (1989): “Applied Logistic Regression.” New
York John Wiley, pp 1-307.
Hynd G, Semrud-Clikeman M (1989a): Dyslexia and brain morphol-
ogy. Psychol Bull 106:447-482.
Hynd G, Semrud-Clikeman M (198913): Dyslexia and neurodevelop-
mental pathology: Relationships to cognition, intelligence and
reading skill acquisition. J Learn Disabil22:204-216.
Jastak JF, Jastak S (1985): “The Wide Range Achievement Test-
Revised.” Wilmington, Delaware: Jastak Associates.
Lahey BB, Russo MF, Walker J L (1989):Personality characteristics of
the mothers of children with disruptive behavior disorders. J Consult
Clin Psychol 57512-515.
Lopez RE (1965): Hyperactivity in twins. Can Psychiatr Assoc J 10:
421426.
Mannuzza S, Gittelman R (1984): The adolescent outcome of hyper-
active girls. Psychiatry Res 13:19-29.
Morrison J (1980): Adult psychiatric disorders in parents of hyper-
active children. Am J Psychiatry 137:825-827.
Morrison JR, Stewart MA (1971): A family study of the hyperactive
child syndrome. Biol Psychiatry 3:189-195.
Morrison JR, Stewart MA (1973): The psychiatric status of the legal
families of adopted hyperactive children. Arch Gen Psychiatry 28:
888-891.
OCallaghan E, Gibson T, Colohan HA, Buckley P, Walshe DG, Larkin
C, Waddington J L (1992): Risk of schizophrenia in adults born af-
New Phenotype Definition of ADHD 377
ter obstetric complications and their association with early onset
of illness: A controlled study. Br Med [Clin Res1305:1256-1259.
Orvaschel H, Puig-Antich J (1987): “Schedule for AfTective Disorder
and Schizophrenia for School-Age Children-Epidemiologic 4th
Version.” Ft. Lauderdale: Nova University, Center for Psychological
Study.
Ott J (1991):Genetic linkage analysis under uncertain disease definition.
In Cloninger CR, Begleiter H (eds): “Banbury Report 33: Genetics
and Biology of Alcoholism.” Cold Spring Harbor, New York Cold
Spring Harbor Laboratory Press, pp 327-331.
Rutter M, Korn S, Birch HG (1963): Genetic and environmental
factors in the development of “primary reaction patterns.” Br J SOC
Clin Psychol 2:161-173.
Safer D J (1973): A familial factor in minimal brain dysfunction. Behav
Genet 3:175-186.
Safer DJ, Krager JM (1988): A survey of medication treatment for
hyperactivelinattentive students. JAMA 260:225&2258.
Schachar R, Wachsmuth R (1990): Hyperactivity and parental psy-
chopathology. J Child Psychol Psychiatry 31:381-392.
Spitzer E, Williams JBW (1986): “The Structured Clinical Interview
for DSM-111-R.”New York New York State Psychiatric Institute.
Stewart MA, de Blois CS, Cummings C (1980): Psychiatric disorder in
the parents of hyperactive boys and those with conduct disorder.
J Child Psychol Psychiatry 21:283-292.
Torgersen AM, Kringlen E (1978): Genetic aspects of temperamental
differences in infants. J Am Acad Child Psychiatry 17:433444.
Tsuang MT, Faraone SV, Lyons M J (1993): Identification of the pheno-
type in psychiatric genetics. Eur Arch Psychiatry Clin Neurosci
682:l-12.
Wechsler D (1974): “Manual for the Wechsler Intelligence Scale for
Children-Revised.” New York Psychological Corporation.
Welner Z, Welner A, Stewart M, Palkes H, Wish E (1977): A controlled
study of siblings of hyperactive children. J Nerv Ment Dis 165:
110-117.
Willerman L (1973): Activity level and hyperactivity in twins. Child
Dev 44:28%293.